Massive slowdown with CoronaAO map in refraction color slot

[chopmeister]

I encountered something weird. I wanted to plug in a CoronaAO map in the refraction color slot of a glass material so that I get some non-transparent edges. What happened was pretty weird so I made a little test scene to see if it's somehow related to the scene I was working on - and it is not.

To reproduce just make some boxes or whatever, and make a CoronaMtl with the default settings, only reflection and refraction set to 1, ie. standard transparent glass. Render.
Now put a CoronaAO map in the reflection slot of that material, and render again.

This is my test scene, just a plane, CoronaSun and a few boxes with the glass material.



Here are screenshots from my test:



This is without the CoronaAO map. The 1st pass took approx. 27 seconds, and notice 7 mil rays/s. Now WITH CoronaAO in the refraction slot:



1:21, so three times slower, and check out the stats: 23.2 mil rays/s...

I would expect slowdowns but not by 300% for such a simple scene. Also, what's with the rays/s stats?

In a more complex scene this also impacts UHD cache speeds, where I encountered a 500% percent slower solution (maybe even more) just by adding one box with a material like this.


[EDIT] Forgot to add: I'm using Corona 1.01.00 (Jun 16 2015) with 3dsmax 2016.

[maru]

But you know that you can use Corona 1.1.1.? :)
https://corona-renderer.com/download/


I noticed that there is a big slowdown when you use AO as opacity too, but it renders much much faster if you remove the ground plane. Maybe Corona doesn't like calculating shadows for opacity based on AO?

[chopmeister]

But you know that you can use Corona 1.1.1.? :)
https://corona-renderer.com/download/

Hahaha, yes I know, but sadly didn't have time to reinstall the farm. :)

[Ondra]

the AO shader does some sampling that can get costly, it is expected to lose some performance if you apply it to majority of your scene. I would maybe suggest decreasing number of samples in its settings (I think we changed defaults there already to make it faster by default)